A tradeoff between reliability and dimensionality in the neural code for music

The rich experience of listening to music depends on the neural integration of its constituent elements within the early auditory pathway. Here, we performed the first large-scale study of neural responses to complex music to characterize the neural coding of individual instruments and mixtures with both normal hearing and mild-to-moderate hearing loss. Using coherence and manifold analyses along with deep learning, we identified strong nonlinear interactions in mixed music that impacted the fidelity and geometry of the neural code, with reliability and dimensionality varying inversely across music types. This tradeoff was also evident in responses to speech and persisted for music even after hearing loss, the effects of which were largely corrected with suitable amplification. These results suggest that the neural coding of music involves an inherent tradeoff and highlight a fundamental challenge in maintaining fidelity while processing sounds with varying spectrotemporal properties.